Fix memory leaks in pyVaultNodeOperationCallback.

[Hoikas]

A long standing issue has been mysterious "key only" memory leaks when the plResManager destructs, as seen in the leak report in plasmadbg.log:

(10/09 07:00:12) 	Leaks in page Personal>psnlMYSTII[000d0022]:
(10/09 07:00:12) 		plSceneObject: (S:0xd0022 F:0 I:526 N:ImgrPhotoPlane02 C:[0,0]) (key only, 10 refs left)
(10/09 07:00:12) 		plLogicModifier: (S:0xd0022 F:0 I:134 N:cRgnImagerActivate_Enter C:[0,0]) (key only, 1 refs left)
(10/09 07:00:12) 		plLogicModifier: (S:0xd0022 F:0 I:135 N:cRgnImagerActivate_Exit C:[0,0]) (key only, 1 refs left)
(10/09 07:00:12) 		plResponderModifier: (S:0xd0022 F:0 I:336 N:RespImager2Button C:[0,0]) (key only, 1 refs left)
(10/09 07:00:12) 		plPythonFileMod: (S:0xd0022 F:0 I:86 N:PythonFileImager02 C:[0,0]) (key only, 13 refs left)
(10/09 07:00:12) 		plDynamicTextMap: (S:0xd0022 F:0 I:1 N:ImagerDynamic02_dynText C:[0,0]) (key only, 1 refs left)

These leaks correspond to the xSimpleImager.py script, which uses vaultOperation callbacks. The vault operation callback C++ code was incorrectly assuming that PyObject_GetAttrString() would return a borrowed reference and not decrementing the reference count. This is fixed by using pyObjectRef to correct and clarify the reference counting.

Unfortunately, Python still reports errors on shutdown, but this is a good step toward correcting them.

[Hoikas]

I've force-pushed an update that makes the pyVaultNodeOperationCallback constructors explicit to remove potential ambiguity.

[Hoikas]

I need to change pyObjectNewRef to safely accept nullptr, don't merge this yet. Done